Face descriptor of AdaBoost Patterns of Oriented Edge Magnitudes (APOEM) for face recognition

This paper proposes a novel and efficient face descriptor, based on Patterns of Oriented Edge Magnitudes (POEM) features. AdaBoost is exploited to select optimally the most informative and discriminative features in each POEM image. Each face image is first represented by three POEM images, containing various significant features in different regions (sub-windows). The multi-class problem of face recognition is transformed into three independent two-class ones by classifying every two POEM images as intra-personal or extra-personal ones. The χ² distance between corresponding POEM histograms of two POEM images is used as discriminative features. We optimize the parameters of APOEM and apply the whitened principal component analysis (WPCA) dimensionality reduction technique to get a lower dimension and more discriminative face descriptor. Experimental results on FERET and CAS-PEAL-R1 face databases demonstrate that the proposed method achieves better performance than the face descriptors based on WPCA-POEM and POEM.